From at-sensor observation to at-surface reflectance - calibration steps for Earth observation hyperspectral sensors

With the continued development of space borne hyperspectral sensors (CSA HERO, ESA CHRIS-on-PROBA, ESA SPECTRA, NASA SpectraSat) to follow the EO-1 Hyperion sensor, high spectral and spatial Earth observation data will become more readily available to the research and user communities. With this improvement in spectral and spatial resolution comes the need to have more rigorous image preprocessing. Spectral and spatial registration and radiometric response need to be characterized and applied more frequently, possibly on a scene by scene basis depending on the stability of the sensor. This requires a system that can evaluate a dataset and determine these parameters efficiently and independently. A pre-processing procedure to transform at-sensor signals to at-surface reflectance for Earth Observation hyperspectral imagery has been developed at the Canada Centre for Remote Sensing/Natural Resources Canada (CCRS/NRCan). This process examines an image cube for bad pixels (stripes) and noise levels, determines spectral (smile effect) and spatial (keystone) registration per pixel, as well as evaluating the image cube for optimal signal gain and offset, and applies the relevant corrections. Where applicable, a scene-based (vicarious) calibration procedure can also be applied.